News about Gemma 4 E4B

Google DeepMind's Gemma 4 12B enables multimodal AI on laptops with only 16GB RAM, natively handling text, images, and audio while approaching the performance of 26B models. Available on Hugging Face under Apache 2.0. #Gemma4 #MultimodalAI #LocalAI #OpenSource #DeepMind

Google DeepMind released DiffusionGemma — a text diffusion model built by converting Gemma 4 instead of training from scratch, using <10% of original compute. It refines 256-token blocks in parallel, hitting ~1,500 tokens/sec on H100. Two-stage training (reconstruction + SD·RL merging reinforcement learning with sampler distillation) boosts reasoning quality by ~10 points while quadrupling tokens per step. Bidirectional reasoning enables self-correction mid-generation — solving 85% of Sudoku puzzles and fixing math errors before output finalizes. Tradeoffs: lower peak quality than autoregressive Gemma 4, repetition loops from aggressive step reduction, multimodal formatting issues, and throughput parity with standard LLMs at ~32 concurrent users. Released under Apache 2.0 on Hugging Face as experimental foundation for efficient diffusion research. Already powering multilingual speech recognition and interactive radiology reports. #DiffusionGemma #TextDiffusion #GoogleDeepMind #Gemma4 #AIResearch

Google releases Gemma 4, its most capable open model family yet under Apache 2.0 license. Four sizes from 2B to 31B parameters run on devices from smartphones to workstations. Built on Gemini 3 tech, they excel at reasoning, math, coding, and agentic workflows with native function calling and JSON output. The 31B model ranks #3 and 26B MoE #6 on Arena AI leaderboard, outperforming models 20x larger. Available now on Hugging Face, Kaggle, Ollama, and Google platforms with broad framework/hardware support. #Gemma4 #GoogleAI #OpenSourceLLM #Apache2 #AIModels

Google DeepMind released Gemma 4 12B, a multimodal open AI model running locally on laptops with just 16 GB RAM. It processes text, images, and audio natively, nearly matching the 26B model's performance while handling speech recognition, code generation, and multi-minute video analysis. Available on Hugging Face, Ollama, and LM Studio under Apache 2.0 for commercial use. #Gemma4 #GoogleDeepMind #OpenAI #MultimodalAI #LocalLLM

Google silently updated Gemma 4 without version bump, fixing tool-calling bugs and truncated responses. Flash Attention 4 brings 25–70% faster prompt processing and 31% lower latency. The 31B model gains up to 10.1% better agentic reasoning. Image OCR now supports 2.51 MP via max_soft_tokens increase. All model sizes updated, including new 12B. Community criticizes lack of "4.1" naming. #Gemma4 #GoogleAI #LLMUpdate #OpenWeights #AIModels

Google DeepMind released Gemma 4 12B, an open multimodal AI model that runs on laptops with just 16 GB RAM, processing text, images, and audio natively. It nearly matches the performance of the twice‑larger Gemma 4 26B and beats Gemma 3 27B on benchmarks. Available on Hugging Face, Ollama, LM Studio under Apache 2.0. #AI #Gemma4 #Multimodal #OpenSource #LLM

Google has released multi-token prediction (MTP) drafters for its open Gemma 4 AI model family, boosting text generation speeds by up to 3x. Traditional LLMs generate one token at a time, wasting compute waiting for data. MTP uses a small auxiliary model to suggest multiple tokens during idle periods, which the main model verifies in a single pass with no quality loss. The speedup works across smartphones, local computers and cloud setups. MTP drafters are open-source under Apache 2.0 on Hugging Face and Kaggle. Gemma 4, launched in early April 2026, has already passed 60 million downloads. #AI #Google #Gemma4 #MachineLearning #TechNews

Google's Gemma 4 is an open‑source on‑device AI that handles text, images, and audio locally, uses agent skills for Wikipedia, maps, QR codes, and more, runs on phones with as little as 6 GB RAM, is up to 4× faster than prior Gen, and lays the groundwork for Gemini Nano 4. Free AI Edge Gallery app on Android/iOS. #Gemma4 #OnDeviceAI #GoogleAI #EdgeAI #AndroidiOS

Google's Gemma 4 brings free, fully on-device AI to phones, processing text, images, and audio without sending data to the cloud. With agent skills it can use Wikipedia, maps, QR generators, and more. The lightweight E2B/E4B variants run on 6‑8 GB RAM, are up to 4× faster and 60% more battery‑efficient, and are released under Apache 2.0. The free Google AI Edge Gallery app is on Android/iOS, and Gemma 4 lays the groundwork for Gemini Nano 4. #AI #Gemma4 #OnDeviceAI #OpenSource #MobileTech

Google has officially launched Gemma 4, a family of four open-source AI models under the fully permissive Apache 2.0 license. The lineup ranges from efficient E2B and E4B variants for smartphones and edge hardware, to powerful 26B MoE and 31B dense models for workstations. Featuring major upgrades in reasoning, math, and agentic workflows, Gemma 4 outperforms significantly larger competitors and is now available across all major AI frameworks and cloud providers. #Gemma4 #OpenSourceAI #GoogleAI #MachineLearning #AI